Testing and Verification

Testing and Verification

JetSetCraft treats a passing Java compile as intermediate evidence, not final proof. The repository CI regenerates assets, validates contracts, builds the reobfuscated JAR, runs real Forge GameTests, and starts a real dedicated server.

Current alpha.5 Style Library and build acceptance

The 0.4.0-alpha.5 implementation merged through PR #44 passed all 44 real Forge GameTests in each of nine configurations, each build and dedicated-server readiness followed by RCON stop and normal saved-dimensions shutdown. This is 396 successful test executions of 44 distinct tests. The base also passed the fifteen-screen native client audit gate. The Style Library receipt records the exact merge and separates CI results from the independent exact-source local visual reproduction.

The feature is 8bb7109, tested through PR integration 6701db2 and squash-merged as 5bad458. All three have Git tree 878476f60413bfe47c18864a99674c363432ce87, with an empty feature-to-merge tree diff.

Final CI evidence	Result and scope
Base run 41388639847	PASS: clean build, 44 GameTests, dedicated-server readiness and normal save/shutdown, fifteen-screen native audit gate
Provider run 41388639856	PASS in all eight lanes: Alex's Mobs, Alex's Caves, Cataclysm, their combined lane, Aether, Iron's, AoA and the combined adventure lane; each passed build, 44 tests and normal server save/shutdown

Artifact inspection boundary: browser URL policy blocked downloading the CI archives. Artifact metadata is API-reported; CI ZIP/JAR bytes were not independently downloaded or hashed, and CI screenshot pixels were not independently reviewed. Identical CI JAR hashes are not claimed. The fifteen-PNG gallery is a separate local reproduction from the exact source, after the required asset generators and rebuild. It passed all three native audit markers and normal save/shutdown. Its local JAR SHA-256 is a5cea430f3b2930b7bae1d479bfda5ac6bf6e08263d3f01e1858835a26fbd847; this is not a hash of a downloaded CI binary. Source/log/image provenance belongs to the new receipt, not the older alpha.4 gallery.

Nine additional acceptance cases

Alpha.5 marker	Production behavior or model boundary exercised
rush_synergy_rules	One preparation, three-family cap, bounded independent soul pools, shield timing and pause, remix cleanup and relay admission
rush_synergy_control	Real owned actors, refreshing Venom, directed Chill, forward Split, foreign-target exclusion, paused/expired effects and native-state preservation
rush_synergy_recovery	Actual owned-rival defeats, shield absorption, sixth-hit recovery, wall-limited soul burst, nonrecursive secondary damage and normal return
rush_cache_options	Built floor markers, distinct private previews, actual claims/cooldown/magnet, immutable snapshots and truthful partially capped rewards
rush_cache_boundaries	Authored obstacles, blocked approaches, current-room charge, between-room hiding and finished-run rejection
rush_route_exploration	Six built rooms, distinct offered reward families, unvisited profiles, honest history and optional-provider/material resolution
crew_style_shared_rewards	Production shared settlements, pre-house learning, founding transfer, replay denial, deadline-based mission claim and saved-data round trips
crew_style_persistence	Real packet codec/handler, stale/negative/unknown/locked rejection, schema 3, schemas 1/2 migration and exact future-payload preservation
crew_style_entry	Earned provider-gated lessons, actual entry with exactly one starting rank, active-run rejection, unchanged return state and operational-house fallback

Cache and route cases inject recordDefeat() counters and compress clocks; the synergy cases separately settle real session-owned kills through production damage handling. Crew cases supply qualifying transaction inputs and saved mission deadlines rather than replaying complete street games or waiting real mission minutes. Codec/handler checks do not constitute remote-network tests. The receipt preserves these limits.

Historical alpha.4 reward, journey and build acceptance

The 0.4.0-alpha.4 implementation merged through PR #43 passed all 35 real Forge GameTests in each of nine configurations, each configuration's build, and dedicated-server readiness followed by an RCON stop and verified normal saved-dimensions shutdown. The base also completed the graphical audit and produced thirteen reviewed 1280 × 720 native captures. This is 315 successful test executions, not 315 different tests. The reward/journey receipt, machine-readable evidence and gallery record the exact artifacts and limits.

The verified feature is b120db1, tested through PR integration b0506b0 and merged as e826694. All three have Git tree 1aab5454652554bffee78299a8f86a2c19013261. All nine artifacts contain the same 5,644,758-byte JAR, SHA-256 182e4b589191c288472b059e9be71b89374687768f079580301153d97498e24c; their downloaded ZIP digests matched GitHub's artifact records.

Configuration	Completed alpha.4 job
Base Forge 1.20.1 / Java 17	Verify JetSetCraft #286: build, 35 tests, normal server save/shutdown and thirteen reviewed client captures
Alex's Mobs 1.22.9	Provider #59 — alexsmobs: build, 35 tests and normal server save/shutdown
Alex's Caves 2.0.2	Provider #59 — alexscaves: build, 35 tests and normal server save/shutdown
Cataclysm 3.31	Provider #59 — cataclysm: build, 35 tests and normal server save/shutdown
Alex's Mobs + Alex's Caves + Cataclysm	Provider #59 — all: build, 35 tests and normal server save/shutdown
Aether 1.20.1-1.5.2-neoforge	Provider #59 — aether: build, 35 tests and normal server save/shutdown
Iron's Spells 1.20.1-3.4.0.2	Provider #59 — irons_spells: build, 35 tests and normal server save/shutdown
AoA 1.20.1-3.7.1 Beta	Provider #59 — aoa: build, 35 tests and normal server save/shutdown
Aether + Iron's Spells + AoA + Cataclysm	Provider #59 — adventure_all: build, 35 tests and normal server save/shutdown

Seven additional acceptance cases

Alpha.4 marker	Production behavior exercised
crew_progress_history	Actual pre-house rewards, sixteen-row history, saved identity, once-only founding transfer and a separate longer replay guard
crew_progress_caps	Credited amounts after caps, explicit busy/no-focus training results and affordable goals that still require physical readiness
crew_progress_mission	Mission reward goes to its assigned member, with the actual saved level transition, consumed job and replay rejection
rush_journey_preview	Actual room offers, objectives and reward previews; stale/unoffered selection rejection and immutable snapshots
rush_journey_history	Real constructed-room admission, bounded six-room history, supplied settlement records without invented payouts, stale receipt rejection and Encore reset
rush_build_rules	Finite distance charge, genuine aerial observations, exact-rank remix, unavailable-provider and stale-version rejection, and the three-slot limit
rush_build_runtime	Real manager/owned-rival movement and combat paths, visible-target mark, secondary-hit isolation, pause/expiry behavior and exact source-state return

The rank-one Breakbeat test preserves its mark and active clock through five paused manager calls, then proves the mark remains at active tick 49 and expires at exactly 50 active ticks. It checks removal of the visible Exposed label and damage multiplier, no replacement mark hiding the result, and unchanged native mob health/effects. Every configuration recorded JETSETCRAFT_RUSH_MARK_EXPIRY active_ticks=50 paused_calls=5. These fixture-driven production calls do not claim a fifty-tick human maneuver or a human balance result.

The new client frames show the first Journey map, initial Build with empty artifact slots, and Progress with empty reward history and a real next house goal. No completed room, remix, claimed mission, displayed earned receipt or developed build is pictured. Server GameTests supply their respective mechanic and settlement evidence.

Historical alpha.3 activity and arena acceptance

The activity and arena implementation merged in PR #41 passed all 28 real Forge GameTests in each of nine configurations, plus every configuration's build and dedicated-server startup checks. The base also verified normal dedicated-server save/shutdown and produced ten reviewed 1280 × 720 real-client captures. The historical receipt and dated scope correction preserve exact identities and explain that the eight provider smoke jobs used SIGTERM/SIGKILL cleanup without a normal-shutdown assertion. GameTest-server normal-save output remains separate valid evidence; alpha.4's stronger gate does not retroactively change those jobs.

The verified feature commit is c733852, tested through PR integration 2626a8d. PR #41 merged as e57e9ab. All three commits have Git tree e5e127156d0335a9c6f2e33ba153e8ecfc75e757, tying the merged implementation to the tested source. These results were recorded on 2026-11-01.

Configuration	Completed evidence
Base Forge 1.20.1 / Java 17	Verify JetSetCraft #280: build, 28 GameTests, dedicated-server readiness/save/shutdown and ten real-client captures
Alex's Mobs 1.22.9	Provider #56 — alexsmobs: build, 28 GameTests and dedicated-server startup smoke (normal shutdown not asserted)
Alex's Caves 2.0.2	Provider #56 — alexscaves: build, 28 GameTests and dedicated-server startup smoke (normal shutdown not asserted)
Cataclysm 3.31	Provider #56 — cataclysm: build, 28 GameTests and dedicated-server startup smoke (normal shutdown not asserted)
Alex's Mobs + Alex's Caves + Cataclysm	Provider #56 — all: build, 28 GameTests and dedicated-server startup smoke (normal shutdown not asserted)
Aether 1.20.1-1.5.2-neoforge	Provider #56 — aether: build, 28 GameTests and dedicated-server startup smoke (normal shutdown not asserted)
Iron's Spells 1.20.1-3.4.0.2	Provider #56 — irons_spells: build, 28 GameTests and dedicated-server startup smoke (normal shutdown not asserted)
AoA 1.20.1-3.7.1 Beta	Provider #56 — aoa: build, 28 GameTests and dedicated-server startup smoke (normal shutdown not asserted)
Aether + Iron's Spells + AoA + Cataclysm	Provider #56 — adventure_all: build, 28 GameTests and dedicated-server startup smoke (normal shutdown not asserted)

Client-audit follow-up

A post-merge repeat passed build, all 28 GameTests and dedicated-server smoke but stopped after eight client captures when the disposable house fixture's activity handoff could not locate its Boombox. PR #42 supplies a real dry supported room, observes grounded client state and hands over the completed home's authoritative anchor. It preserves ordinary item-use packets, native screenshots and return assertions; only two opt-in audit Java classes changed under src.

Verify JetSetCraft #282 then passed its build, all 28 GameTests, server smoke, ten captures and all three visual-audit PASS markers. The feature fe679cf, tested integration c0b2028 and actual merge 76f44fe share tree ffc85769b8d50143cb7c2e1c66f240ed01515bf8. The complete follow-up receipt records the initial failure and observed room/anchor checks.

The historical alpha.3 JAR remains the exact original PR #41 artifact tested in all nine configurations. PR #42 has its own passing base result; ordinary gameplay sources and resources are identical. That ten-image gallery remains the original inspected capture set, separate from alpha.4.

Activity and arena test coverage

Alpha.3 marker	Scope
activity_progression	Cross-mode pre-house earnings, once-only transfer, training, receipts, account persistence and schema migration
activity_tag	Actual Boombox preview and launch, source actors, owner attack hooks, six objectives and bystander/replay denial
activity_safety	Stable random invitation, stale context rejection, three-hit Showdown, orphan/missing actor and anchor removal handling
activity_hill	Contested capture rejection, physical presence at three hills and 480 real uncontested ticks
activity_dimension_return	Selected provider versions and actual Aether/Precasia → circuit → exact source return, with unchanged Minecraft/ride state
rush_arena_catalogue	Live profile loading, optional gating, fallback materials, authored map bounds and forbidden payload rejection
rush_arena_rules	Six-room completion/Encore, three artifact slots, cache refresh and stale draft denial
rush_arena_session	Real manager route choices, generated passable connection, actual owned rivals, shared settlement and exact return
rush_spell_relay	Provider-absent behavior, or the real native Iron Firebolt collision, bounded damage, ownership, virtual defeat and cleanup

The route rules/session tests use explicit fixture-clock compression for long arena durations. They do not speed up the real-client audit. The graphical audit shows a real Boombox activity preview and a genuine three-choice first route, selects that route, enters the connected-arena travel phase and returns normally. Arena combat, objective completion and settlement are established by GameTests; the client audit does not show a cleared room or a reward screen.

The provider-present proofs passed with the actual source mods loaded. The Iron's and adventure_all jobs report native Firebolt admission, collision, bounded virtual damage, foreign-owner rejection and cleanup. The Aether/AoA jobs report actual aether:the_aether and aoa3:precasia departure and exact source return; the combined job reports both dimensions. Provider-absent configurations separately pass the clean unavailable paths. Full native spellbooks, arbitrary foreign capabilities, every modded mob and long human sessions remain outside these results. See Compatibility for the precise adapter contracts and known upstream messages.

Versioned verification evidence

The published v0.3.0 release has nine GameTests, as recorded in its release evidence. The historical 0.4.0-alpha.1 checkpoint added six Rival Rush tests, for fifteen; its receipt and original captures are preserved. The historical 0.4.0-alpha.2 source added four crew-house/integration tests, bringing that checkpoint to nineteen. Alpha.3 added nine activity/arena tests, reaching 28. Alpha.4 added seven reward/journey/build tests, reaching 35. Alpha.5 adds nine Style Library, synergy and cache/route cases, reaching 44, passing in all nine configurations above. Development evidence does not change the published stable release's version or history.

Historical alpha.2 checkpoint

The recorded alpha.2 candidate is d5ecadd, tested through PR integration b9bcec8. PR #40 merged as b852dfc, whose Git tree is identical to that tested integration. The crew-house verification receipt records those source identities, the binary, workflow artifacts and captures.

Configuration	Completed evidence
Base Forge 1.20.1 / Java 17	Verify JetSetCraft #274: clean build, all nineteen GameTests, dedicated-server readiness and normal RCON shutdown, and the eight-screen real Minecraft client audit
Alex's Mobs, Alex's Caves, Cataclysm, and all three together	Provider matrix #51: four separate jobs, each passing its build, all nineteen GameTests and dedicated-server smoke

These links identify that checkpoint's evidence. Later code changes require their own passing run. Provider jobs establish the documented registry and gameplay regression baseline; they do not prove every optional creature's behavior or every combination of other mods.

Local source checks

python -m pip install -r tools/requirements.txt
python tools/generate_models.py
python tools/generate_animations.py
python tools/generate_brand.py
python tools/validate_assets.py
python tools/validate_gameplay_contract.py
python tools/validate_premium_polish.py
python tools/validate_java_syntax.py
python tools/validate_wiki.py
./gradlew clean build --no-daemon
./gradlew runGameTestServer --no-daemon

The model generator is offline-safe: when the pinned optional graffiti archive cannot be fetched, it removes any partial download and produces the original JetSetCraft fallback catalog.

Forge GameTests

The nine original tests verify:

1. Hoverboard registry identity, persistent loadout NBT, and the production ground movement solver.
2. Neutral ride input cannot relaunch stale momentum, deliberate input restarts correctly, directional sprint kickoff works while neutral kickoff cannot manufacture motion, and swimming retains the exact vanilla velocity vector.
3. Scooter registry identity, persistent loadout NBT, and the universal JetSetMovement path.
4. Breakdance starting with no ride gear, server-owned move selection, completed-phrase scoring, and Flow.
5. Immediate dance cancellation when a combat swing takes presentation/input authority.
6. Stable one-to-one identity for all 28 dance moves, complete 24-trick catalog coverage, and hostile input/NBT sanitization.
7. Same-entity Street Gear identity, physical-item persistence, real entity NBT save/load, clean removal, spider/slime anatomy routing, persistent gangification/de-gangification, and a real placed Boombox proving Zombie Head → Dead Beat tuning plus start/cancel/immediate-restart/target-return behavior.
8. All sixteen registered/craftable paint-balloon colors, deterministic bounded splat encoding, Tag/Free Paint item state, real 1,000-ray surface placement, exact decal materialization, and persistent six-face sizing.
9. Built-in and custom graffiti NBT persistence, including the support position, plus automatic removal after the supporting wall is broken. The current fixture observes a complete supported cycle first, then requires native discard within 40 additional entity ticks; all nine configurations recorded native_support_ticks=40.

The six Rival Rush tests introduced in alpha.1 remain part of the current suite:

10. Deterministic three-choice upgrades, rejection of stale or repeated choices, bounded ranks and counters, the thirty-minute decision pause, and explicit Endless continuation.
11. All seven mob-card mastery thresholds, increasing milestone rewards, three starting-draft unlocks, distinct gang/species identities, saved album data, and unchanged Minecraft XP.
12. Recovery through the production return-journal path, exact original inventory and ride NBT, survival state and effects, preserved album progress, and rejection of incomplete, foreign-owner, nonfinite, or duplicate-slot journals before inventory mutation.
13. Actual session entry and exit across dimensions, loan equipment, exact original position/items/ride/XP restoration, and Forge capability invalidation/revival with the same durable ride data.
14. Real director-created actors and card pickups, independent player ownership, rejection of cross-session damage and duplicate claims, and suppression of event XP without changing unrelated Minecraft mob XP.
15. The real spray-can placement hook and points for a fresh wall, rejection of repaint farming, and authoritative upgrade actions that reject stale run IDs, stale offer versions, and replays.

The four alpha.2 crew-house/integration tests require these exact CI pass-marker IDs:

Test marker	Production behavior exercised
crew_house	A real placed Boombox founds one house; physical facility registration and support-dependent upgrades charge the correct supplies; expansion preserves placed blocks, checks overlap before spending and acquires no forced chunks.
crew_missions	Dispatch freezes role-adjusted rewards and a saved deadline; stale, early, repeated and cross-player claims are rejected; the exact ready-time boundary settles supplies, member/crew XP and reputation once; cancellation grants no reward.
crew_persistence	World SavedData round-trips member/job identity, facilities, progress and run receipts; unknown schemas remain read-only; missing anchors retain the crew; relocation clears physical bindings while keeping purchased tiers; duplicate Rush settlement and training of busy members are rejected.
crew_rush	Actual house-to-Rush entry grants one chosen preparation alongside an independent earned album draft; the director creates real owned rivals, production tagging defeats them, and the real return path awards bounded crew progress once while preserving ordinary inventory/XP.

Mission deadline checks use the persisted ready-time boundary deterministically. The integration test advances only its run's active clock before exercising actual actors and return; it does not claim a one-minute human play session or alter the shared world clock.

The player tests use uniquely identified Forge fake players and a real ServerLevel rather than mocked movement classes. Unique profiles prevent parallel GameTests from moving or mutating the same automation player. Isolated activity, arena and graffiti fixtures explicitly supply their required terrain and bounded temporary loading/ticking tickets; those fixture tickets do not add production chunk-loading behavior.

FakePlayer tests establish in-memory lifecycle behavior; they do not establish player-file durability. The real-client audit exercises the actual save acknowledgement and network entry/return paths. Forced process termination, storage failure, arbitrary third-party player capabilities, and a human thirty-minute or long Endless playtest are separate acceptance work; no passing gate above implies those have been exercised.

Input transport acceptance

Client controls remain change-driven: any digital mask or analog movement change is transmitted immediately. While riding, dancing, or holding live movement input, unchanged state is refreshed every 5 client ticks instead of being resent every tick. A neutral inactive client sends its release state once and then stays quiet. The server independently sanitizes unknown bits/non-finite axes and clears stale armed input after 20 ticks, so the active heartbeat retains a four-times safety margin without needless 20-packets-per-second steady-state traffic.

Dedicated-server smoke

CI launches Forge with eula=true, waits for the real ready line, and fails on lifecycle/mod-loading/fatal JetSetCraft errors. The base and all eight provider jobs issue an RCON stop and require normal saved-dimensions shutdown output and successful process completion. Alpha.5 passed this gate, introduced in alpha.4. JARs and logs are uploaded as workflow artifacts; the current CI archives remain uninspected under the access limitation above. Older provider startup-only smoke results retain their historical scope.

Known upstream messages retained in passing logs

The historical alpha.3 adventure_all log contains two nonfatal upstream issues alongside its successful tests and dedicated-server startup smoke:

• AoA's remote update response: Forge's version checker could not parse the response from the URL declared in AoA's META-INF/mods.toml (Expected ':' at line 120 column 13). This affected update information, not the tested arena or dimension data.
• Iron's concentration-amulet recipe: data/irons_spellbooks/recipes/curios/concentration_amulet.json referenced unregistered irons_spellbooks:mithril_ingot, so Minecraft rejected recipe irons_spellbooks:curios/concentration_amulet. That crafting recipe did not load.

The final alpha.4 combined adventure job still records the rejected Iron recipe alongside all 35 passing tests and normal dedicated-server save/shutdown. The AoA checker parse failure was not observed in that final job's GameTest or dedicated-server logs; it remains a historical observation.

The Compatibility warning ledger identifies the exact resources and evidence. Passing integration results do not certify all provider recipes or features, and the upstream resources and version pins were not modified to conceal these messages.

Real-client visual audit

Run gradlew.bat -Djetsetcraft.visualAudit=true runClient --no-daemon on Windows (or the wrapper equivalent elsewhere). The opt-in harness quick-loads its development world, creates a controlled inline-skate/HUD/painted-wall scene, captures the selector in persistent Free Paint mode and the custom painter, prints JETSETCRAFT_VISUAL_AUDIT_PASS, and exits cleanly. It is inert in ordinary launches.

The development CI also runs ./gradlew runClient -Djetsetcraft.rushVisualAudit=true --no-daemon with a disposable RivalRushAudit world prepared by the build workflow. It enters and leaves Rush through real network actions, compares the returned dimension, inventory and vanilla XP with its observed pre-entry fixture, and writes the setup, district HUD, empty-album and return-result captures.

After return, the crew audit uses an actual Charter on placed fixture blocks, founds a home through production actions, checks two real roster profiles, registers an operational workshop and dispatches a real timed job. It captures Home, Crew, Facilities and Jobs. The activity audit adds Activities, Arena route, Journey map, Build and Crew Progress; alpha.5 adds Cache Scout and Style Library, bringing the gate to fifteen 1280 × 720 captures. It requires the Rush, crew and activity VISUAL_AUDIT_PASS markers, observed fixtures and all fifteen capture markers/files. The final CI gate passed; direct CI-pixel inspection remains unavailable.

For Cache Scout, the harness chooses an actual offered route, then explicitly logs a disposable-world fixture that teleports to its selected arena and loads nearby chunks. The production builder, arrival rules and cache loop supply the displayed state; the fixture does not fabricate clocks, defeats, rewards or clears. The audit requires zero completed arenas and records the six live caches. The empty-progression Style Library fixture shows 0 crew XP, 0/16 learned and HOME selected, with three lessons on page 1 of 6. These are initial-state inspections, not human traversal, an earned cache, unlocked-style or completed-circuit demonstrations. The alpha.5 receipt and local gallery identify this independent exact-source reproduction.

Historical alpha.4 pixel review: the thirteen-screen reward/journey gallery remains that checkpoint's original reviewed evidence. The ten-screen alpha.3 gallery and eight-screen alpha.2 gallery retain their own source identities.

In those alpha.4 captures, the album is empty when capture 03 is taken; the later return result in capture 04 shows score 95, one rival and one collected card. Captures 05–07 retain native hover overlays, and the job is still running in capture 08. Captures 09–11 show the Boombox chooser and first route/map; the audit enters connected travel and returns. Capture 12 has zero artifacts. Capture 13 has empty reward history, with 16/40 Supplies toward the 25 × 25 house goal. Forge tests cover actual combat/clear/settlement, receipts, artifact behavior and job claims. That visual audit does not complete a room, claim a mission or demonstrate a populated history. Short automated audits do not measure house progression balance, thirty-minute/Endless enjoyment, sustained performance, physical follower behavior or a complete human session.

Runtime acceptance world

Run:

• /jetsetcraft build_vanilla_lab
• /jetsetcraft status
• /jetsetcraft set_momentum <speed> as an operator-only diagnostic

Then exercise every ride style, dance family, surface lane, rail type, wall, hazard, transfer, combat overlay, dimension transition, death/respawn, reconnect, and multiplayer tracking path.

Wiki verification

tools/validate_wiki.py ensures the complete page set exists, sidebar destinations resolve, internal wiki links are valid, required technical terms are present, and no unfinished-marker language ships. The validated wiki/ directory is the canonical source published to the GitHub wiki.

Premium offline gates

The premium gate additionally rejects malformed or degenerate OBJ geometry, unresolved materials/textures, non-finite or visibly unclosed animation loops, hard optional-mod class imports, invasive mixin/ASM paths, stale client-cache regressions, and Java syntax failures before ForgeGradle begins. These checks are strong preflight evidence, but they never replace the real Forge build, GameTests, dedicated-server smoke, or client acceptance.